Upload File

reproducing the feature outputs of common programs in matlab using melfcc

  • Home
  • Documents
  • Reproducing the Feature Outputs of Common Programs in Matlab Using Melfcc
4
25/3/2015 Reproducing the feature outputs of common programs in Matlab using melfcc.m http://www.ee.columbia.edu/ln/rosa/matlab/rastamat/mfccs.html 1/4 Dan Ellis : Resources : Matlab : PLP, Rasta, MFCC : Reproducing the feature outputs of common programs using Matlab and melfcc.m When I decided to implement my own version of warpedfrequency cepstral features (such as MFCC) in Matlab, I wanted to be able to duplicate the output of the common programs used for these features, as well as to be able to invert the outputs of those programs. This page gives some examples of how cepstra can be calculated by three common programs (HTK 's HCopy, feacalc from SPRACHcore , and mfcc.m from Malcolm Slaney's Auditory Toolbox for Matlab), and how to duplicate the results (or very nearly) using my melfcc.m routine. This also automatically shows you how to invert cepstra calculated by either path into spectrograms or waveforms using invmelfcc.m, since its arguments are the same. HTK MFCC 20130226: For an emulation of HTK's MFCC calculation accurate to the 3rd decimal place, see the modified rastamat code in calc_mfcc . The main differences were that HTK applies preemphasis independently on each window, and also removes the mean on each window. Calculating features in HTK is done via HCopy, which can convert between a wide range of representations including waveform to cepstra. HCopy takes its options from a config file. Thus, to convert 16 kHz sampled soundfiles to standard Melfrequency cepstral coefficients (MFCCs), you would have a file config.mfcc containing: SOURCEKIND = WAVEFORM SOURCEFORMAT = WAVE SOURCERATE = 625 TARGETKIND = MFCC_0 TARGETRATE = 100000.0 WINDOWSIZE = 250000.0 USEHAMMING = T PREEMCOEF = 0.97 NUMCHANS = 20 CEPLIFTER = 22 NUMCEPS = 12 (The SOURCEFORMAT option specifies that the wavefiles are in MSWAVE format.) Then to calculate the features, you simply run HCopy from the Unix command line: $ HCopy ‐C config.mfcc sa1.wav sa1‐mfcc.htk We can emulate this processing in Matlab, and compare the results, as below: (Note that the ">>" at the start of each line is an image, so you can cut and copy multiple lines of text directly into Matlab without having to worry about the prompts). % Load a speech waveform [d,sr] = wavread('sa1.wav '); % Calculate HTK‐style MFCCs mfc = melfcc(d, sr, 'lifterexp', ‐22, 'nbands', 20, ... 'dcttype', 3, 'maxfreq',8000, 'fbtype', 'htkmel', 'sumpower', 0); % Load the features from HCopy and compare: htkmfc = readhtk ('sa1‐mfcc.htk'); % Reorder and scale to be like mefcc output htkmfc = 2*htkmfc(:, [13 [1:12]])'; % (melfcc.m is 2x HCopy because it deals in power, not magnitude, spectra) subplot(311) imagesc(htkmfc); axis xy; colorbar title('HTK MFCC'); subplot(312) imagesc(mfc); axis xy; colorbar title('melfcc MFCC'); subplot(313) imagesc(htkmfc ‐ mfc); axis xy; colorbar title('difference HTK ‐ melfcc'); % Difference occasionally peaks at as much as a few percent (unexplained), % but is basically negligable % Invert the HTK features back to waveform, auditory spectrogram, % regular spectrogram (same args as melfcc()) [dr,aspec,spec] = invmelfcc(htkmfc, sr, 'lifterexp', ‐22, 'nbands', 20, ... 'dcttype', 3, 'maxfreq',8000, 'fbtype', 'htkmel', 'sumpower', 0); subplot(311) imagesc(10*log10(spec)); axis xy; colorbar title('Short‐time power spectrum inverted from HTK MFCCs') subplot(312) specgram(dr,512,sr); colorbar title('Spectrogram of reconstructed (noise‐excited) waveform');

Upload: trunggana-abdul-w

Post on 16-Nov-2015

20 views

Category:

Documents


6 download

Report

DESCRIPTION

matlab instrumentation

TRANSCRIPT

  • 25/3/2015 ReproducingthefeatureoutputsofcommonprogramsinMatlabusingmelfcc.m

    http://www.ee.columbia.edu/ln/rosa/matlab/rastamat/mfccs.html 1/4

    DanEllis:Resources:Matlab:PLP,Rasta,MFCC:

    ReproducingthefeatureoutputsofcommonprogramsusingMatlabandmelfcc.m

    WhenIdecidedtoimplementmyownversionofwarpedfrequencycepstralfeatures(suchasMFCC)inMatlab,Iwantedtobeabletoduplicatetheoutputofthecommonprogramsusedforthesefeatures,aswellastobeabletoinverttheoutputsofthoseprograms.Thispagegivessomeexamplesofhowcepstracanbecalculatedbythreecommonprograms(HTK'sHCopy,feacalcfromSPRACHcore,andmfcc.mfromMalcolmSlaney'sAuditoryToolboxforMatlab),andhowtoduplicatetheresults(orverynearly)usingmymelfcc.mroutine.Thisalsoautomaticallyshowsyouhowtoinvertcepstracalculatedbyeitherpathintospectrogramsorwaveformsusinginvmelfcc.m,sinceitsargumentsarethesame.

    HTKMFCC

    20130226:ForanemulationofHTK'sMFCCcalculationaccuratetothe3rddecimalplace,seethemodifiedrastamatcodeincalc_mfcc.ThemaindifferenceswerethatHTKappliespreemphasisindependentlyoneachwindow,andalsoremovesthemeanoneachwindow.

    CalculatingfeaturesinHTKisdoneviaHCopy,whichcanconvertbetweenawiderangeofrepresentationsincludingwaveformtocepstra.HCopytakesitsoptionsfromaconfigfile.Thus,toconvert16kHzsampledsoundfilestostandardMelfrequencycepstralcoefficients(MFCCs),youwouldhaveafileconfig.mfcccontaining:

    SOURCEKIND=WAVEFORMSOURCEFORMAT=WAVESOURCERATE=625TARGETKIND=MFCC_0TARGETRATE=100000.0WINDOWSIZE=250000.0USEHAMMING=TPREEMCOEF=0.97NUMCHANS=20CEPLIFTER=22NUMCEPS=12

    (TheSOURCEFORMAToptionspecifiesthatthewavefilesareinMSWAVEformat.)Thentocalculatethefeatures,yousimplyrunHCopyfromtheUnixcommandline:

    $HCopyCconfig.mfccsa1.wavsa1mfcc.htk

    WecanemulatethisprocessinginMatlab,andcomparetheresults,asbelow:(Notethatthe">>"atthestartofeachlineisanimage,soyoucancutandcopymultiplelinesoftextdirectlyintoMatlabwithouthavingtoworryabouttheprompts).

    %Loadaspeechwaveform[d,sr]=wavread('sa1.wav');%CalculateHTKstyleMFCCsmfc=melfcc(d,sr,'lifterexp',22,'nbands',20,...

    'dcttype',3,'maxfreq',8000,'fbtype','htkmel','sumpower',0);%LoadthefeaturesfromHCopyandcompare:htkmfc=readhtk('sa1mfcc.htk');%Reorderandscaletobelikemefccoutputhtkmfc=2*htkmfc(:,[13[1:12]])';%(melfcc.mis2xHCopybecauseitdealsinpower,notmagnitude,spectra)subplot(311)imagesc(htkmfc);axisxy;colorbartitle('HTKMFCC');subplot(312)imagesc(mfc);axisxy;colorbartitle('melfccMFCC');subplot(313)imagesc(htkmfcmfc);axisxy;colorbartitle('differenceHTKmelfcc');%Differenceoccasionallypeaksatasmuchasafewpercent(unexplained),%butisbasicallynegligable

    %InverttheHTKfeaturesbacktowaveform,auditoryspectrogram,%regularspectrogram(sameargsasmelfcc())[dr,aspec,spec]=invmelfcc(htkmfc,sr,'lifterexp',22,'nbands',20,...

    'dcttype',3,'maxfreq',8000,'fbtype','htkmel','sumpower',0);subplot(311)imagesc(10*log10(spec));axisxy;colorbartitle('ShorttimepowerspectruminvertedfromHTKMFCCs')subplot(312)specgram(dr,512,sr);colorbartitle('Spectrogramofreconstructed(noiseexcited)waveform');

    http://www.ee.columbia.edu/~dpwe/resources/matlab/rastamat/http://www.ee.columbia.edu/~dpwe/resources/matlab/rastamat/http://rvl4.ecn.purdue.edu/~malcolm/interval/1998-010/http://www.ee.columbia.edu/~dpwe/resources/http://www.ee.columbia.edu/ln/rosa/matlab/rastamat/config.mfcchttp://www.ee.columbia.edu/ln/rosa/matlab/rastamat/readhtk.mhttp://www.icsi.berkeley.edu/~dpwe/projects/sprach/sprachcore.htmlhttp://www.ee.columbia.edu/~dpwe/resources/matlab/http://htk.eng.cam.ac.uk/http://www.ee.columbia.edu/ln/rosa/matlab/rastamat/sa1.wavhttp://www.ee.columbia.edu/~dpwe/http://labrosa.ee.columbia.edu/projects/calc_mfcc/

  • 25/3/2015 ReproducingthefeatureoutputsofcommonprogramsinMatlabusingmelfcc.m

    http://www.ee.columbia.edu/ln/rosa/matlab/rastamat/mfccs.html 2/4

    subplot(313)specgram(d,512,sr);colorbartitle('Originalsignalspectrogram');%Spectrogramslookprettyclose,althoughnoiseexcitation%ofreconstructiongivesitaweird'whisperingcrowd'sound

    HTKPLP

    HTKcanalsocalculatePLPfeatures.ItturnsoutthatthesearesomewhatdifferentfromtheMFCCfeaturesbecausethecepstraarecalculatedbyadifferentalgorithm.However,wecanstillemulateandinvertthemwithdifferentparameters.TocalculatePLPfeatureswithHCopy,weneedanewconfigfile,config.plp:

    SOURCEKIND=WAVEFORMSOURCEFORMAT=WAVESOURCERATE=625TARGETKIND=PLP_0TARGETRATE=100000.0WINDOWSIZE=250000.0USEHAMMING=TPREEMCOEF=0.97NUMCHANS=20CEPLIFTER=22NUMCEPS=12USEPOWER=TLPCORDER=12

    (TARGETKINDischanged,andUSEPOWERandLPCORDERareadded).Thenwecalculatethefeatures:

    $HCopyCconfig.plpsa1.wavsa1plp.htk

    ..andcomparetotheMatlabversion:

    [d,sr]=wavread('sa1.wav');%CalculateHTKstylePLPsplp=melfcc(d,sr,'lifterexp',22,'nbands',20,...

    'dcttype',1,'maxfreq',8000,'fbtype','htkmel',...'modelorder',12,'usecmp',1);

    %LoadtheHCopyfeatureshtkplp=readhtk('sa1plp.htk');%Reorder(noscalinginthiscase)htkplp=htkplp(:,[13[1:12]])';subplot(311)imagesc(htkplp);axisxy;colorbartitle('HTKPLP');subplot(312)imagesc(plp);axisxy;colorbartitle('melfccPLP');subplot(313)imagesc(htkplpplp);axisxy;colorbartitle('differenceHTKmelfcc');%Unexplaineddifferencescanbeupto20%forhigherorder%cepstra,butessentiallythesame

    %InverttheHTKfeaturesbackagainbymirroringargstomelfcc[dr,aspec,spec]=invmelfcc(htkplp,sr,'lifterexp',22,'nbands',20,...

    'dcttype',1,'maxfreq',8000,'fbtype','htkmel',...'modelorder',12,'usecmp',1);

    subplot(311)imagesc(10*log10(spec));axisxy;colorbartitle('ShorttimepowerspectruminvertedfromHTKPLPs')subplot(312)specgram(dr,512,sr);colorbartitle('Spectrogramofreconstructed(noiseexcited)waveform');subplot(313)specgram(d,512,sr);colorbartitle('Originalsignalspectrogram');%Prettyclose

    feacalcMFCC

    http://www.ee.columbia.edu/ln/rosa/matlab/rastamat/config.plphttp://www.ee.columbia.edu/ln/rosa/matlab/rastamat/sa1.wavhttp://www.ee.columbia.edu/ln/rosa/matlab/rastamat/readhtk.m

  • 25/3/2015 ReproducingthefeatureoutputsofcommonprogramsinMatlabusingmelfcc.m

    http://www.ee.columbia.edu/ln/rosa/matlab/rastamat/mfccs.html 3/4

    feacalcisthemainfeaturecalculationprogramfromICSI'sSPRACHcorepackage.It'sactuallyawrapperaroundtheolderrasta.whichwastheoriginalClanguageimplementationofRASTAandPLPfeaturecalculation.feacalchasbeenexpandedtobeabletocalculate(itsownversionof)MFCCfeatures,sotoparalleltheHTKexamplesabove,we'llstartwithfeacalc'sMFCCfeature.Theycanbecalculatedwiththefollowingcommandline:

    $feacalcsr16000nyq8000delta0rasnoplpno\domcepcomnofrqmelfilttricep13opfhtk\sa1.wavosa1fcmfc.htk

    andweduplicatethisinMatlabasfollows:

    [d,sr]=wavread('sa1.wav');%CalculateFeacalcstyleMFCCs%(scaletomatchnormalizationofMelfilters)mfc2=melfcc(d*5.5289,sr,'lifterexp',0.6,'nbands',19,...

    'dcttype',4,'maxfreq',8000,'fbtype','fcmel','preemph',0);%LoadtheHCopyfeaturesfcmfc=readhtk('sa1fcmfc.htk');%Noneedtoreorderorscale,justtransposefcmfc=fcmfc';subplot(311)imagesc(fcmfc(2:13,:));axisxy;colorbartitle('feacalcMFCC');subplot(312)imagesc(mfc2(2:13,:));axisxy;colorbartitle('melfccMFCC(feacalcstyle)');subplot(313)imagesc(fcmfcmfc2);axisxy;colorbartitle('differencefeacalcmelfcc');%Smalldifferencesinhighordercepstradueto%cumulativeerrorsinMelfiltershapes

    ..andinvertingworksjustthesameasabove.

    feacalcPLP

    feacalcwasoriginallydesignedtocalculatePLP(andRasta)features,sothisisitsmore'native'invocation:

    $feacalcsr16000nyq8000delta0rasnodomcepplp12\opfhtksa1.wavosa1fcplp.htk

    ..whichweduplicatethisinMatlabasfollows:

    [d,sr]=wavread('sa1.wav');%CalculateFeacalcstylePLPsplp2=melfcc(d,sr,'lifterexp',0.6,'nbands',21,...

    'dcttype',1,'maxfreq',8000,'fbtype','bark','preemph',0,...'numcep',13,'modelorder',12,'usecmp',1);

    %LoadtheHCopyfeaturesfcplp=readhtk('sa1fcplp.htk');%justtransposefcplp=fcplp';subplot(311)imagesc(fcplp(2:13,:));axisxy;colorbartitle('feacalcPLP');subplot(312)imagesc(plp2(2:13,:));axisxy;colorbartitle('melfccPLP(feacalcstyle)');subplot(313)imagesc(fcplpplp2);axisxy;colorbartitle('differencefeacalcmelfcc');%Afewlocalizeddifferencesduewindowsetc.

    ..andonceagaininvertingworksjustthesameasabove.

    AuditoryToolboxmfcc.m

    ThemostpopulartoolforcalculatingMFCCsinMatlabismfcc.mfromMalcolmSlaney'sAuditoryToolbox.ThisiswhatIusedforalongtime,untilIneededsomethingwithmoreflexibility.Thatflexibilityincludesbeingabletoduplicatemfcc.m.Here'showwecancomparetheminMatlab.

    [d,sr]=wavread('sa1.wav');%CalculateMFCCsusingmfcc.mfromtheAuditoryToolbox%(gainshouldbe2^15becausemelfccscalesbythatamount,%butinthiscasemfccuses2xFFTlen)ce=mfcc(d*(2^14),sr);%Scalethemtomatch(log_10andpower)ce=log(10)*2*ce;

    http://www.ee.columbia.edu/ln/rosa/matlab/rastamat/sa1.wavhttp://www.icsi.berkeley.edu/~dpwe/projects/sprach/sprachcore.htmlhttp://www.ee.columbia.edu/ln/rosa/matlab/rastamat/sa1.wavhttp://www.ee.columbia.edu/ln/rosa/matlab/rastamat/readhtk.mhttp://www.ee.columbia.edu/ln/rosa/matlab/rastamat/sa1.wavhttp://www.ee.columbia.edu/ln/rosa/matlab/rastamat/readhtk.m

  • 25/3/2015 ReproducingthefeatureoutputsofcommonprogramsinMatlabusingmelfcc.m

    http://www.ee.columbia.edu/ln/rosa/matlab/rastamat/mfccs.html 4/4

    %Duplicatewithmelfcc.mmfc3=melfcc(d,sr,'lifterexp',0,'minfreq',133.33,...

    'maxfreq',6855.6,'wintime',0.016,'sumpower',0);%..andcompare:subplot(311)imagesc(ce(2:13,:));axisxy;colorbartitle('AuditoryToolboxMFCC');subplot(312)imagesc(mfc3(2:13,:));axisxy;colorbartitle('melfccMFCC(AudToolboxstyle)');subplot(313)imagesc(cemfc3);axisxy;colorbartitle('differenceAudTBoxmelfcc');%Smalldifferencesmainlyduetohanningvs.hamming

    NotesonthedifferencesbetweendifferentMFCCs

    MelmappingfunctionMelfilternormalizationDCTusedtocalculatecepstrumNumberofMelbands(andhencetheirwidth)FrequencyspanofMelbandsLifteringrasta,htk,noneDetailsofinitialSTFT(odd/evenhann/hamm,fftlength,windowlength)MelintegrationinlinearorpowerdomainDitherandDCremovalPreemphasis

    Lastupdated:$Date:2013/02/2617:00:16$

    DanEllis

    http://www.ee.columbia.edu/~dpwe/mailto:[email protected]


Self Reproducing Machines

Theory of Self-reproducing Automata

Chapter 8 Self-reproducing computer programs Chapter 8: Self-reproducing programs 1

Deploying MATLAB Applications in Excel, Java, and .NET ...€¦ · • Overview of MathWorks Services . 3 Technical Computing Workflow Reporting and Documentation Outputs for Design

What’s New in MATLAB and Simulink · Using MATLAB and Simulink to Build Deep Learning Models Inputs Input Design Design Outputs Output Data Machine Learning Deep Learning Model

Uncertainty Quantification for Multiphysics Applicationsneckel/siamuq16_slides_minisymp/2016_psuade... · PSUADE provides rich Matlab/Scilab graphics outputs and many data (sample)

An Example of a Reproducing Kernel Hilbert Space · 2019. 2. 2. · turns out also a reproducing kernel Hilbert space. Keywords Isoperimetric inequality · Hilbert spaces · Reproducing

Reproducing kernel Hilbert spaces and regularizationspaces ... · Reproducing kernel Hilbert spaces and regularizationspaces and regularization ... H is a vector space, ... the reproducing

Reproducing Empire

Prospects of Reproducing Useful Transparencies

REPRODUCING THE RECORDED ARTS IN THE ZONE pic/hifiplus/ISS46... · REPRODUCING THE RECORDED ARTS Issue 46 £4.50 REPRODUCING THE RECORDED ARTS MAXX Power - from Wilson Audio Continuous

Reproducing the Observed Universe with Simulations

Introduction to MATLAB - Marine Biological Laboratory...You can cut and paste commands in this packet directly into MATLAB. Outputs that you will see from MATLAB are shown in black

Satellite Attitude Control System Design Using … · Simplorer and Matlab. Aerospace Power & Electronics Simulation Workshop 2004 29 Matlab/Simulink Model Outputs from the Simplorer

AN 314 Digital Predistortion Reference Design · MATLAB GUI Altera provides a bit-accurate and cycle-accurate (cycle-accurate only at outputs) MATLAB model of the DPD reference design

Using Verilog for Testbenches - syssec.ethz.ch outputs to expected outputs and report errors. Carnegie Mellon 13 Testbench with Testvectors ... Can be done in C, Perl, Python, Matlab

Reproducing Emily Dickinson’s Furniture

What’s New in MATLAB and Simulink€¦ · Using MATLAB and Simulink for Reinforcement Learning Inputs Design Design Outputs Outputs Machine Learning Deep Learning Model Simulink

On self-reproducing computer programs

Cartridge Tape Recording and Reproducing

MAJLUFF. Reproducing Nations Catalogue

Enacting the International/ Reproducing Eurocentrism

GANmera: Reproducing Aesthetically Pleasing Photographs ...openaccess.thecvf.com/content_CVPRW_2019/papers/... · GANmera: Reproducing Aesthetically Pleasing Photographs using Deep

REPRODUCING KERNEL HILBERT SPACES - Bilkent … · of reproducing kernel Hilbert spaces, generation of new spaces and relationships ... The Hilbert space with reproducing kernel K

Electronic Home Music Reproducing Equipment

Reproducing Graphs Chris Cannings & Richard Southwell

Reproducing Kernels in Hilbert Spaces - math.unl.edujorr1/classes/2009summer/math896/... · Reproducing Kernels in Hilbert Spaces Thomas Clark August 10 - 11, 2009. Reproducing Kernels

NAB AUDIO RECORDING AND REPRODUCING ......NAB AUDIO RECORDING AND REPRODUCING STANDARDS FOR AUDIO CASSETTE TAPE SYSTEMS The NAB Recording and Reproducing Standards Committee w,as originally

Applications of Reproducing Kernel Hilbert

Dynamic Power Allocation for Standalone PV -micro grid ... · MATLAB SIMULINK model is utilized to analyze the working proposing multilevel Supercapacitor HESS.Simulation outputs

Gilat 5th Edition, Chapter 4 Fall 2017 MATLAB Assignment C: OUTPUTSrlan/math152_F2017/M152C.pdf · Fall 2017 MATLAB Assignment C: OUTPUTS Math 152-5xx, Team A, #0 Your Name, UIN 123456789

THEORY OF REPRODUCING KERNELS(')Reproducing kernels of finite-dimensional classes. 346 4. Completion of incomplete Hubert spaces. 347 5. The restriction of a reproducing kernel. 350

Information recording method, reproducing method and reproducing

NAB AUDIO RECORDING AND REPRODUCING STANDARDS FOR DISC RECORDING … · NAB AUDIO RECORDING AND REPRODUCING STANDARDS FOR DISC RECORDING AND REPRODUCING including A GLOSSARY OF TERMS

Reproducing Kernel Hilbert Spaces